Bedrock ConverseStream mid-stream retryable errors unmarshalled to unknown exception

[steffeng]

When ConverseStream experiences throttling during the streaming of the response, the error codes are returned camelCase (instead of PascalCase) like serviceUnavailableException, throttlingException.

The ConverseStream docs tell:

The exception returned is wrapped with CompletionException, so you need to invoke Throwable.getCause() to retrieve the underlying exception.

Unfortunately, the SDK does not deal with the camelCase and unmarshals to a BedrockRuntimeException ("Unknown exceptions will be thrown as an instance of this type.") instead of the known and documented ServiceUnavailableException, or ThrottlingException, making this a an unexpected and special case for retry/ exception handling. See the attached debug screenshot:

Bedrock Runtime team refused to fix the error code on their end in the response post-launch. Can you handle camelCase on the SDK side for a uniform handling of the exception mid-stream and outside of stream, so retry mechanisms work on capacity and throttling issues?

Regression Issue

• [ ] Select this option if this issue appears to be a regression.

Expected Behavior

All mid-stream errors are unmarshalled to their corresponding exception types (e.g. throttlingException -> ThrottlingException).

Current Behavior

All mid-stream errors are all unmarshalled to BedrockRuntimeException.

Reproduction Steps

Run many converseStream requests in parallel on a busy day/ time.

Possible Solution

No response

Additional Information/Context

No response

AWS Java SDK version used

2.28.0

JDK version used

Corretto-21.0.8.9.

Operating System and version

MacOS 26.1